Where is morality on wheels? Decoding large language model (LLM)-driven decision in the ethical dilemmas of autonomous vehicles
Published in Travel Behaviour and Society, 40, 101039, 2025
Summary
Large Language Models (LLMs) are increasingly used in autonomous driving, yet their ethical decision-making remains under-explored. This study proposes a framework to evaluate LLM ethics using a stated preference survey with 13,122 dilemma scenarios involving factors like group size, age, and risk. Analyzing GPT-4 series and Mistral-Large models via binary logit and decision trees reveals significant differences in preferences (e.g., prioritizing pedestrians) and sensitivities to factors like estimated deaths, age, and gender.
Links
DOI: 10.1016/j.tbs.2025.101039
